The Simulations And Researches On Data-driven Modeling Methods

Modeling is the important basis for the control and optimization of industrial process. The nature of industrial processes can be understood clearly through mechanism modelling, but building mechanism models often require a large amount of prior knowledge and practical experience, and the basic rules of the actual system. Statistical modeling usually relates to the input and output data without requiring internal mechanism of the proees, so the model can apply to non-linear and uncertain systems.This thesis focuses on the modeling methods on data-driven, and simulates and researches on the continuous heating furnace and the Tennessee-Yisiman chemical industry process. The main work completed is summarized as follows:Firstly, combining with the data of the furnace and the data of the Tennessee-Eastman process simulation, the BP Neural Network model is established with the two ways.Secondly, according to the way of constructing the data, the Least Squares Support Vector Machines(LSSVM) is built on the data of the furnace and the Tennessee-Eastman process simulation. Experiment show that the accurcy of the LSSVM model is better than that of the BP Neural Network model.Thirdly, a combination of modeling methods with the Kernel Independent Component Analysis and the Least Squares Support Vector Machines is proposed, KICA-LSSVM for short, on the furnace and the the Tennessee-Eastman process.The results of the study proved the effectiveness of the KICA-LSSVM model, and finally, the summary and perspectives of the KICA-LSSVM model are addressed.